Grok 3 represents a substantial advancement over Grok 2, boasting several new features and capabilities. These improvements include enhanced logical reasoning and coding, with early tests showing a 20% improvement in coding accuracy. Grok 3 also prioritizes bias reduction by incorporating broader datasets, aiming for a fairer AI model and introduces updated system prompts addressing political bias directly. Furthermore, it expands into multimodal capabilities with image analysis and Aurora, a proprietary text-to-image generation tool. A future voice mode is also under development, promising improved accessibility. Grok 3’s self-improvement capabilities allow it to monitor its outputs and learn from its mistakes. These advancements are underpinned by a significant increase in computational power, having been trained using an extensive infrastructure with approximately 100,000 GPUs and 200 million GPU hours, a tenfold increase compared to Grok 2.